Data-Driven Control

Data-driven control studies how to choose actions from measured system responses, either without an explicit model or with a model learned from data. Common approaches include behavioral methods such as Willems' fundamental lemma and data-enabled predictive control, Koopman and reproducing-kernel operator models, neural dynamics models inside model predictive control, and reinforcement learning or direct policy learning. These methods are applied to robots, vehicles, traffic, and energy systems, often combining predictive optimization with physical structure, passivity, Lyapunov constraints, or safety filters. A central challenge is balancing data efficiency and adaptation to changing conditions with reliable stability, constraint satisfaction, and performance beyond the training data.